tanbul 2004
A METHOD OF IMAGE RESOLUTION ENHANCEMENT BASED ON
THE MATCHING TECHNIQUE
Pingxiang Li, Huanfeng Shen, Liangpei Zhang
State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University,
Wuhan 430079, China. (pxli64@163.com)
KEY WORDS: Photogrammetry, Image, Processing, Resolution, Software, Matching.
ABSTRACT:
In the field of digital photogrammetry, it is very important to enhance the image resolution. By enhancement, a clearer image with
higher resolution is produced. So far, the enhancement technique is widely applied in various photogrammetric images. However,
because of the restriction of the CCD sensor itself, the number of pixels on the sensor isn’t much enough in some case. The image
quality is affected and restricted. To solve this problem, the enhancement techniques are expended mainly in two categories: One is
hardware solution; the other is software solution. In this paper, we propose a software algorithm for the enhancement of the image
resolution considering inaccurate sub-pixel matching. In the proposed algorithm, the shifts, the gray values of the low-resolution
images and the enhancement ratio are used to calculate the gray values of the higher- resolution image iteratively. Thus, the new
image has higher resolution, so that it has higher definition. Experimental results indicate that the proposed algorithm has more
universal applications.
1. INTRODUCTION
In digital photogrammetry imaging application, images with
high spatial resolution are desired. Generally, high-resolution
images are obtained depends on hardware solution, that it to
say they are obtained directly from high precision optics and
charge coupled devices (CCDs). However, the cost for high-
precision optics and sensors is not inappropriate for general-
purpose commercial applications, and the technology of CCDs
and high precision optics cannot keep up with the demand for
high-resolution images due to technical limits of sensor
dimensions, shot noise etc. As a result, many software
algorithms have been designed to obtain a high-resolution
image.
Recently, it has been one of the most active research areas to
enhance a high-resolution image from a number of low-
resolution frames of the same scene. The topic of it has
received considerable attention in research community. Early
research on it dates back to the work by Huang and Tsai in
1984 (Tsai, 1984). They solved the problem in the frequency
domain. Since then, researchers, primarily within the
engineering community, have worked out many kinds of
algorithms, such as non-uniform interpolation approach (Clark,
1985), projection onto convex sets (POCS) approach (Stark,
1989; Tekalp, 1992), stochastic approach (MAP estimate
approach (Schulz, 1996) and ML estimate approach (Tom,
1995)), iterative back-projection (IBP) approach (Irani, 1991),
adaptive filtering approach (Elad, 1999) etc.
In 2001, Fryer John and Kerry McIntosh presented a rigorous
geometric (RG) algorithm to enhance a higher resolution image
from several overlapping, and slightly offset, images of low
resolution based on image matching technique (Fryer, 2001).
This algorithm may be utilized for applications successfully
where a higher resolution is desired than has been achieved
previously. However, the algorithm is sensitive to noise in the
input images, and the matching error wasn't considered. To
overcome the limitations, this paper proposed a matching-
error-considered extension of Fryer RG algorithm.
This paper is organized as follows: in section II, the rigorous
geometric algorithm of Fryer is overviewed; in section III, the
influence of the matching error on the enhancement process is
analyzed; in section IV, the extended algorithm is proposed.
Experimental results are shown in section V, and we conclude
in section VI.
2. TRIGOROUS GEOMETRIC(RG) ALGORITHM
The steps of RG algorithm can be expressed as follows(Fryer,
2001):
I. Collect several low-resolution images, and select an
enhancement ratio (range 1.1 to 1.9).
2. Select an image as reference arbitrarily, and
determine pixel offsets of each other image from the
reference image.
3. Form sets of equations using the offsets as
coefficients, the enhancement ratio, and the grey
values of the low-resolution images as observations.
4. Solve the sets of equations for higher resolution
pixels using least squares.
5. Display the resultant higher resolution image.
The most important step of this algorithm is how to form the
sets of equations in step 3. In the equations formation, the
geometric relationships between coarse and fine pixels must be
used. To develop the relationships, each pixel in the coarse
images must be “mapped” onto the fine pixels coordinate
system, thus determining which fine or unknown pixels are
affected by each individual coarse pixel. For example in Fig. 1,
the coarse pixel C2 covers the area bound by from (0.5, 2) to (2,
3.5) in the fine pixel coordinate system.
These coordinates show the upper, lower, left and right bounds
of the coarse pixel. Using these bounds, the proportion of the
